Inverse Design of Potential Dicyanamide Anion Based Ionic Liquids with High Nernst–Einstein Conductivity

Ionic liquids exhibit wide electrochemical windows, negligible volatility, and strong design flexibility, making them promising electrolytes for next-generation energy storage devices. However, the vast design space renders conventional trial-and-error discovery inefficient. Here, we develop a reinforcement learning framework for the de novo inverse design of potential dicyanamide-anion-based ionic liquids with enhanced Nernst–Einstein conductivity. A recurrent neural network trained on known cation structures was coupled with Monte Carlo tree search to generate novel cations. A forward predictive machine learning model was implemented as the reward function, guiding the search toward candidates with high Nernst–Einstein conductivity. Generated structures were subsequently evaluated using molecular dynamics simulations, and the resulting Nernst–Einstein conductivity data were iteratively incorporated to retrain and refine the forward model, forming a closed reinforcement learning loop. Through five iterations, the maximum molecular-dynamics-simulated Nernst–Einstein conductivity of the generated potential ionic liquids increased from 0.61 S/m to 3.77 S/m. The liquid-state character and the synthesizability of the generated candidates remain to be established. The proposed framework could facilitate the computational screening of ionic liquids with desired properties.